| name | tech-design-review |
| description | Reviews the latest committed tech-design docs on a branch and sends an advisory report back to the requester session. |
Tech-Design Review
Review committed tech-design docs and return an advisory report.
Workflow protocol baseline: use the agent-deck-workflow skill.
Input
Provide one of:
- the mailbox body from
tech_design_review_requested
- direct scope + committed design docs + base branch + problem/goals/constraints
Direct-use mode is valid.
In mailbox mode, treat the body as a review brief and committed-doc pointer, not as the full design source.
Inspect the latest committed docs on the stated branch before judging the design.
Review Dimensions
Review the requested tech-design snapshot using these dimensions:
- problem framing and scope
- constraints, assumptions, and success criteria
- decision quality and tradeoffs
- simplicity, coupling, and change surface
- compatibility, migration, and rollback
- operational readiness, observability, and failure handling
- security, privacy, and data boundary risks
- unresolved questions that block implementation confidence
This is not code review.
Architect Role
Review this like a senior engineer in a mature production environment:
- skeptical of hand-wavy claims
- protective of compatibility and operational clarity
- intolerant of unclear ownership, hidden coupling, and weak migration stories
Inspection Order
- Validate the problem and scope first.
- Is this solving a real problem?
- Are non-goals, constraints, and success conditions clear enough to judge the proposal?
- Review the proposed design through the dimensions above.
- Focus first on decision quality, coupling, and change surface.
- Then review migration, operations, and security.
- Distinguish missing evidence from design failure.
- A weakly justified claim is not automatically a wrong design, but it is still a reviewable gap.
- Call out when the proposal direction looks sound but the support, constraints, or rollout detail is too thin.
- Produce one advisory report.
- Prioritize the few objections or caveats most likely to change implementation confidence.
Required Baseline
Before reviewing quality, verify:
- problem statement is stated
- tech-design branch is stated
- tech-design base branch is stated
- in-scope design docs are stated
- alternatives or rejected options are stated
- major constraints are stated
Hard block:
- if problem statement or in-scope design docs are missing, ask one short clarification question in direct-use mode
- in mailbox mode, continue but mark the report as
NEEDS_REVISION and list the missing critical items under Major Risks
Soft gaps:
- if tech-design base branch, alternatives/rejected options, or major constraints are missing, continue the review
- record the missing items under
Design Gaps
If critical context is still missing after one clarification in direct-use mode:
- mark the report as
NEEDS_REVISION
- list the missing items under
Major Risks
Output Format
Mailbox mode uses the full structure below:
Task: <task_id>
Action: tech_design_review_report
From: architect <architect_session_id>
To: <requester_role> <requester_session_id>
Round: <round>
## Summary
[One-line architect summary]
## Reviewed Scope
- Base branch: [branch the tech-design branch started from]
- Branch: [tech-design branch]
- Commit: [reviewed branch HEAD]
- Docs reviewed:
- `path/to/doc1.md`
- `path/to/doc2.md`
## Decision
SOUND | SOUND_WITH_CAVEATS | NEEDS_REVISION
## Major Risks
- [risk or `None`]
## Design Gaps
- [gap or `None`]
## Alternatives
- [alternative or `None`]
## Questions To Resolve
- [question or `None`]
## What Looks Good
- [strength or `None`]
## Tool Context
- Architect tool profile: [architect_tool_profile or `explicit`]
- Architect tool cmd: [architect_tool_cmd]
## Residual Risk
[What remains uncertain after this review]
Decision guidance:
SOUND: the design is coherent and implementation-ready with no material blockers
SOUND_WITH_CAVEATS: the core direction is sound and deliverable, but non-blocking caveats or tracked follow-up work should be recorded before or during implementation
NEEDS_REVISION: the current design is missing critical framing, contains a material flaw, or is too incomplete to trust as the implementation basis
Direct-Use Mode
When invoked directly by the user instead of mailbox workflow:
- skip the mailbox header block
- keep the same review sections starting at
## Summary
- return the report directly in the conversation
Agent Deck Mode
Use the agent-deck-workflow skill for shared protocol.
Skill-specific context resolution:
task_id: explicit -> mailbox body -> ask
architect_session_id: explicit -> mailbox body To header -> bound mailbox sender context -> ask
requester_session_id: explicit -> mailbox body From header -> ask
requester_role: explicit -> mailbox body From header -> default requester
tech_design_base_branch: explicit -> mailbox body Base branch -> ask
architect_tool_cmd: explicit -> mailbox body Tool Context; if missing, use the shared tool-resolution contract for role architect
architect_tool_profile: explicit -> mailbox body Tool Context -> resolver tool_profile
round: explicit -> mailbox body Round header -> default 1
Execution flow:
- preserve or resolve
architect_tool_profile / architect_tool_cmd
- review the latest committed tech-design docs on the referenced branch
- for round
>1 in the same architect session, compare current branch HEAD against the previous reviewed commit from the prior report/mailbox context
- if the previous reviewed commit is unavailable, use git history for the in-scope docs and state the baseline uncertainty under
Residual Risk
- produce one
tech_design_review_report
Rules
- architect is review-only in this lane
- review docs and design rationale only
- keep feedback advisory, skeptical, and evidence-based
- do not use
SOUND_WITH_CAVEATS for unresolved blocking issues; use NEEDS_REVISION when the next doc revision is required before the design is deliverable
- do not treat your own feedback as final authority; the requester may disagree and argue the design tradeoff
- ask the user to decide when the disagreement becomes subjective, strategic, or stuck
- prefer concrete design objections over generic taste comments
- focus on the highest-leverage risks first: wrong problem, bad tradeoff, hidden coupling, broken migration, weak operational story
- acknowledge design strengths that should be preserved during implementation, especially when they create useful constraints or simplify the solution space